Brain = Google ?

A microcosm of all that's good and bad about science in the media.



I just came across an article at New Scientist talking about research that compares our brains to Google's page rank procedure. The idea of the brain being googlesque is quite catchy, the kind of thing you could drop into conversation and seem quite intelligent for doing so. But before we go spreading that meme, it's worth taking a closer look. When we scratch the surface, things get a bit murky, although hopefully my scratching around clears some things up too.



It's only a short article, so I might as well let it speak for itself, with my self speaking for me in between. I will deal with the ideas as they are presented, but the problem is that without reading the journal article it's about, who's really to say what's going on? Very few readers will ever take the time to read the original journal article, and I'd bet that very few writers do either - at least to the level that allows a bit of perspective and criticism. This issue is absolutely ubiquitous in science writing, but I don't have time to address it now. Suffice to say it places a huge burden on the science writer's measly shoulders. Anyway....



Our memory for words can be modelled as a network in which each point represents a different word, with each linked to words that relate to it.



Ok, so we're talking about a model. Here in a nutshell is the daily focal point of most scientists - a model. It stands before them, naked and exposed to the world, doing its best to represent reality. Of course, scientists are equally (hopefully more, maybe less) focused on the reality their model is modelling too. In this case, we're trying understand... what? Something about the way our mind uses words. So let's build a model. It is a clearly fallacious to represent a word as a solid, defined entity in its own right, but we'll forgive them that because all models must necessarily simplify somewhere. Still, I can't help wondering whether they're putting the cart before horse in already specifying which words are related to which. Isn't how the mind super-specially figures this out for us a mystery worth investigating?



Psychologist Tom Griffiths and colleagues at the University of California, Berkeley, wondered whether the ease with which the brain retrieves words is similar to the way that websites are ranked by PageRank: by the number of sites that link to them.



It's essential for a science writer to mention the author, and preferably host institution, of the research. That way interested folk are just a hop, click and jump from finding the original research article, or the researcher's general interests, or commercial interests, or whatever.



I couldn't help but wondering just where Google popped up in all of this. Did it drive the research? Did it emerge unbidden from the pure experimental results? Was it a cute rejoinder in the original paper's discussion, aimed at lightening the tone? Was it added by the researcher or someone from their uni's PR department to get the mercilessly fickle attention of media Monitors in science mags? Anyway, it seems like a reasonable hypothesis - the more words that "link to" a given word, be it in terms of spelling, etymology, the sound of the word or whatever, the easier the word will be to retrieve. The more hits it will get!


It seems it might. In tests against other word-retrieval algorithms, PageRank most clearly matched the human model (Psychological Science, vol 18, p 1069).



Ok, so the crucial missing ingredient is some, any kind of description of a) the other word-retrieval algorithms, and b) some perspective as to how big a deal this is in the field. Was it unexpected? Was there another model that previously occupied the position of best model, that has now been ceremoniously (publishing results is a ceremony of sorts!) dumped in favour of the PageRank model of human word retrieval? Just what the Dickens are the implications of all this??



The results suggest human memory studies could be improved by examining the tricks that search engines employ, and vice versa, says Griffiths.



Ok, there they are. Just who is going to follow this new approach? The authors? Anyone else? Ultimately, they rely on the validity of the analogy between human brains and computers/internets.